Fwd: DMA problems in ide-scsi

Fwd: DMA problems in ide-scsi

[meaty biscuit]

I know there are lots of people that are glad to be done with
ide-scsi, but I'm hoping there is someone out there that has some
experience with this driver that my be able to help.  I would happily
switch modules and start using ide-cd, but I have a few pieces of
software that rely on ide-scsi to work properly and I don't have
enough time to change my software to work with ide-cd before my
product release deadline.

I am working with a mainline kernel, version 2.6.15.7 (I cannot change
kernel versions either).  If DMA is enabled and I try to write to a
CD, I get a kernel panic.  However, reading from a CD with DMA enabled
works fine.  If DMA is disabled and programmed IO is used, I can both
read and write CDs but the fact that PIO uses so much of the CPU
causes my application to have some problems and again, I don't have
time to go through several application release cycles to make them
work with PIO.

I have noticed that writing to CD (with DMA enabled) in 2.6.9 works
fine, it seems as though the breakage of ide-scsi occured in 2.6.10.
Also, burning a CD using DMA with ide-scsi in 2.6.19 seems to work as
well.  I have looked through the ide-scsi code for hours, and I have
also done a fair amount of debugging looking for the problem but I
have had no success.  I tried contacting Bartlomiej and have been
unsuccessful in getting a hold of him.  Does anyone know of a patch
floating around that may fix this problem?  Does anyone that is more
familiar with the ide, scsi, or dma subsystems have any suggestions
for me?  I am willing to put in the time and effort to fix this
problem and I would be more than happy to submit a fix back into the
open source world, but I am stuck and need any help I can get.

Thanks for your time.

Re: DMA problems in ide-scsi

[Bartlomiej Zolnierkiewicz]

Hi,

On 1/11/07, meaty biscuit <meatybiscuit@gmail.com> wrote:
> I know there are lots of people that are glad to be done with
> ide-scsi, but I'm hoping there is someone out there that has some
> experience with this driver that my be able to help.  I would happily
> switch modules and start using ide-cd, but I have a few pieces of
> software that rely on ide-scsi to work properly and I don't have
> enough time to change my software to work with ide-cd before my
> product release deadline.
>
> I am working with a mainline kernel, version 2.6.15.7 (I cannot change
> kernel versions either).  If DMA is enabled and I try to write to a
> CD, I get a kernel panic.  However, reading from a CD with DMA enabled
> works fine.  If DMA is disabled and programmed IO is used, I can both
> read and write CDs but the fact that PIO uses so much of the CPU
> causes my application to have some problems and again, I don't have
> time to go through several application release cycles to make them
> work with PIO.
>
> I have noticed that writing to CD (with DMA enabled) in 2.6.9 works
> fine, it seems as though the breakage of ide-scsi occured in 2.6.10.
> Also, burning a CD using DMA with ide-scsi in 2.6.19 seems to work as

If it works fine in the current kernels it seems like the problem
is not kernel bug but the fact that you are stuck on 2.6.15.7.

> well.  I have looked through the ide-scsi code for hours, and I have
> also done a fair amount of debugging looking for the problem but I
> have had no success.  I tried contacting Bartlomiej and have been
> unsuccessful in getting a hold of him.  Does anyone know of a patch

I've just read your mail from Jan 4 (was on TODO).
[ sorry but ide-scsi problems are (very) low priority ]

> floating around that may fix this problem?  Does anyone that is more
> familiar with the ide, scsi, or dma subsystems have any suggestions
> for me?  I am willing to put in the time and effort to fix this
> problem and I would be more than happy to submit a fix back into the
> open source world, but I am stuck and need any help I can get.

If you send me the log/photo of the kernel panic I might be able to help you.

Bart

Re: DMA problems in ide-scsi

[meaty biscuit]

Here is a log of the kernel panic (via serial console, if there is a
better way to capture the output please let me know):

ide-scsi: hdb: suc 91
ide-scsi: hdb: que 92, cmd = [ 51 0 0 0 0 0 0 0 64 0 ]
ide-scsi: hdb: suc 92, rst = [ 0 20 0 1 1 1 1 0 ff 0 0 0 0 0 0 0 ]
ide-scsi: hdb: que 93, cmd = [ 52 1 0 0 0 1 0 0 ff 0 ]
ide-scsi: hdb: suc 93, rst = [ 0 22 1 1 0 6 42 1 0 0 0 0 0 0 0 0 ]
ide-scsi: hdb: que 94, cmd = [ 52 1 0 0 0 1 0 0 ff 0 ]
ide-scsi: hdb: suc 94, rst = [ 0 22 1 1 0 6 42 1 0 0 0 0 0 0 0 0 ]
ide-scsi: hdb: que 95, cmd = [ 0 0 0 0 0 0 ]
ide-scsi: hdb: suc 95
ide-scsi: hdb: que 96, cmd = [ 2a 0 0 0 0 0 0 0 80 0 ]
Unable to handle kernel NULL pointer dereference at virtual address 00000024
 printing eip:
f88c0969
*pde = 00000000
Oops: 0000 [#1]
PREEMPT SMP
Modules linked in: sg sr_mod cdrom snd_pcm_oss snd_mixer_oss nvidia
agpgart i2c_core snd_hda_intel snd_hda_codec snd_intel8x0
snd_ac97_codec snd_ac97_bus snd_pcm snd_timer snd soundcore
snd_page_alloc UniMc nfsd lockd nfs_acl sunrpc usbserial e1000
af_packet iptable_filter ip_tables ide_scsi aic7xxx scsi_transport_spi
piix ide_core psmouse microcode ext3 jbd mbcache usbmouse usbkbd
usbhid uhci_hcd ehci_hcd usbcore rtc xfs exportfs sd_mod ata_piix
libata scsi_mod
CPU:    0
EIP:    0060:[<f88c0969>]    Tainted: P      VLI
EFLAGS: 00010086   (2.6.15.7-gemOS-v009)
EIP is at idescsi_expiry+0x1b/0x44 [ide_scsi]
eax: 00000000   ebx: f883e2e0   ecx: 00000001   edx: 00000286
esi: f8a47c58   edi: f88c0992   ebp: ffffffff   esp: c02fdee4
ds: 007b   es: 007b   ss: 0068
Process swapper (pid: 0, threadinfo=c02fc000 task=c02b5b40)
Stack: f88c2240 00000286 f7495e80 f7495e80 f8a33cdd f8a47c58 00000102 f882d18c
       c19070e0 00000286 f7495e80 00000102 f8a33c90 c19070e0 c0127a0b f7495e80
       a56b6b80 000f422a c02b5b40 c02fc000 c02fdf34 c02fdf34 c02b5b40 c02f3308
Call Trace:
 [<f8a33cdd>] ide_timer_expiry+0x4d/0x1d2 [ide_core]
 [<f882d18c>] scsi_times_out+0x0/0x8c [scsi_mod]
 [<f8a33c90>] ide_timer_expiry+0x0/0x1d2 [ide_core]
 [<c0127a0b>] run_timer_softirq+0xdf/0x1d2
 [<c01234e3>] __do_softirq+0xbf/0xd2
 [<c012353e>] do_softirq+0x48/0x4a
 [<c0123648>] irq_exit+0x42/0x44
 [<c0103aac>] apic_timer_interrupt+0x1c/0x24
 [<c0100ff8>] mwait_idle+0x4a/0x54
 [<c0100fae>] mwait_idle+0x0/0x54
 [<c0100e18>] cpu_idle+0x82/0xa0
 [<c02fe8e7>] start_kernel+0x189/0x1c6
 [<c02fe328>] unknown_bootoption+0x0/0x1a8
Code: a5 89 c1 83 e1 03 74 02 f3 a4 0f b6 45 00 eb 97 83 ec 10 89 5c
24 0c 8b 44 24 14 8b 40 1c 8b 58 10 c7 04 24 40 22 8c f8 8b 43 2c <8b>
40 24 89 44 24 04 a1 00 34 2f c0 89 44 24 08 e8 3a db 85 c7
 <0>Kernel panic - not syncing: Fatal exception in interrupt
ide-scsi: abort called for 96

This is the output when the module is compiled with debug messages
turned on, notice the last SCSI command:  ide-scsi: hdb: que 96, cmd =
[ 2a 0 0 0 0 0 0 0 80 0 ]

I checked the SCSI standard, 2a is the "write (10)" command.

Below is another log that I captured when attempting to burn a CD
(again with DMA enabled) using a different application (internally
developed) that writes to CD by sending SCSI ioctls directly to the
/dev/sgX device (as opposed to burning an ISO or another file using
cdrecord):


ide-scsi: The scsi wants to send us more data than expected - discarding data
ide-scsi: transferred 1 of 2 bytes
ide-scsi: The scsi wants to send us more data than expected - discarding data
ide-scsi: transferred 1 of 2 bytes
hdb: DMA timeout retry
hdb: timeout waiting for DMA
ide-scsi: No active request in idescsi_eh_reset
hdb: status timeout: status=0xd0 { Busy }
ide: failed opcode was: unknown
hdb: drive not ready for command
hdb: status timeout: status=0xd0 { Busy }
ide: failed opcode was: unknown
hdb: drive not ready for command
ata1: BUG: timeout without command

Thanks again to anyone that can respond.  Special thanks to Bart!



On 1/10/07, Bartlomiej Zolnierkiewicz <bzolnier@gmail.com> wrote:
> Hi,
>
> On 1/11/07, meaty biscuit <meatybiscuit@gmail.com> wrote:
> > I know there are lots of people that are glad to be done with
> > ide-scsi, but I'm hoping there is someone out there that has some
> > experience with this driver that my be able to help.  I would happily
> > switch modules and start using ide-cd, but I have a few pieces of
> > software that rely on ide-scsi to work properly and I don't have
> > enough time to change my software to work with ide-cd before my
> > product release deadline.
> >
> > I am working with a mainline kernel, version 2.6.15.7 (I cannot change
> > kernel versions either).  If DMA is enabled and I try to write to a
> > CD, I get a kernel panic.  However, reading from a CD with DMA enabled
> > works fine.  If DMA is disabled and programmed IO is used, I can both
> > read and write CDs but the fact that PIO uses so much of the CPU
> > causes my application to have some problems and again, I don't have
> > time to go through several application release cycles to make them
> > work with PIO.
> >
> > I have noticed that writing to CD (with DMA enabled) in 2.6.9 works
> > fine, it seems as though the breakage of ide-scsi occured in 2.6.10.
> > Also, burning a CD using DMA with ide-scsi in 2.6.19 seems to work as
>
> If it works fine in the current kernels it seems like the problem
> is not kernel bug but the fact that you are stuck on 2.6.15.7.
>
> > well.  I have looked through the ide-scsi code for hours, and I have
> > also done a fair amount of debugging looking for the problem but I
> > have had no success.  I tried contacting Bartlomiej and have been
> > unsuccessful in getting a hold of him.  Does anyone know of a patch
>
> I've just read your mail from Jan 4 (was on TODO).
> [ sorry but ide-scsi problems are (very) low priority ]
>
> > floating around that may fix this problem?  Does anyone that is more
> > familiar with the ide, scsi, or dma subsystems have any suggestions
> > for me?  I am willing to put in the time and effort to fix this
> > problem and I would be more than happy to submit a fix back into the
> > open source world, but I am stuck and need any help I can get.
>
> If you send me the log/photo of the kernel panic I might be able to help you.
>
> Bart
>